Search

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting., 27 pages, 17 figures + references and appendices, repo: https://github.com/google/BIG-bench

We show that a GPT-3 model can learn to express uncertainty about its own answers in natural language -- without use of model logits. When given a question, the model generates both an answer and a level of confidence (e.g. "90% confidence" or "high confidence"). These levels map to probabilities that are well calibrated. The model also remains moderately calibrated under distribution shift, and is sensitive to uncertainty in its own answers, rather than imitating human examples. To our knowledge, this is the first time a model has been shown to express calibrated uncertainty about its own answers in natural language. For testing calibration, we introduce the CalibratedMath suite of tasks. We compare the calibration of uncertainty expressed in words ("verbalized probability") to uncertainty extracted from model logits. Both kinds of uncertainty are capable of generalizing calibration under distribution shift. We also provide evidence that GPT-3's ability to generalize calibration depends on pre-trained latent representations that correlate with epistemic uncertainty over its answers., CalibratedMath tasks and evaluation code are available at https://github.com/sylinrl/CalibratedMath

Aims To determine the impact of COVID-19 on orthopaediatric admissions and fracture clinics within a regional integrated care system (ICS). Methods A retrospective review was performed for all paediatric orthopaedic patients admitted across the region during the recent lockdown period (24 March 2020 to 10 May 2020) and the same period in 2019. Age, sex, mechanism, anatomical region, and treatment modality were compared, as were fracture clinic attendances within the receiving regional major trauma centre (MTC) between the two periods. Results Paediatric trauma admissions across the region fell by 33% (197 vs 132) with a proportional increase to 59% (n = 78) of admissions to the MTC during lockdown compared with 28.4% in 2019 (N = 56). There was a reduction in manipulation under anaesthetic (p = 0.015) and the use of Kirschner wires (K-wires) (p = 0.040) between the two time periods. The median time to surgery remained one day in both (2019 IQR 0 to 2; 2020 IQR 1 to 1). Supracondylar fractures were the most common reason for fracture clinic attendance (17.3%, n = 19) with a proportional increase of 108.4% vs 2019 (2019 n = 20; 2020 n = 19) (p = 0.007). While upper limb injuries and falls from play apparatus, equipment, or height remained the most common indications for admission, there was a reduction in sports injuries (p < 0.001) but an increase in lacerations (p = 0.031). Fracture clinic management changed with 67% (n = 40) of follow-up appointments via telephone and 69% (n = 65) of patients requiring cast immobilization treated with a 3M Soft Cast, enabling self-removal. The safeguarding team saw a 22% reduction in referrals (2019: n = 41, 2020: n = 32). Conclusion During this viral pandemic, the number of trauma cases decreased with a change in the mechanism of injury, median age of presentation, and an increase in referrals to the regional MTC. Adaptions in standard practice led to fewer MUA, and K-wire procedures being performed, more supracondylar fractures managed through clinic and an increase in the use of removable cast. Cite this article: Bone Joint Open 2020;1-7:424–430.

In many contexts, lying -- the use of verbal falsehoods to deceive -- is harmful. While lying has traditionally been a human affair, AI systems that make sophisticated verbal statements are becoming increasingly prevalent. This raises the question of how we should limit the harm caused by AI "lies" (i.e. falsehoods that are actively selected for). Human truthfulness is governed by social norms and by laws (against defamation, perjury, and fraud). Differences between AI and humans present an opportunity to have more precise standards of truthfulness for AI, and to have these standards rise over time. This could provide significant benefits to public epistemics and the economy, and mitigate risks of worst-case AI futures. Establishing norms or laws of AI truthfulness will require significant work to: (1) identify clear truthfulness standards; (2) create institutions that can judge adherence to those standards; and (3) develop AI systems that are robustly truthful. Our initial proposals for these areas include: (1) a standard of avoiding "negligent falsehoods" (a generalisation of lies that is easier to assess); (2) institutions to evaluate AI systems before and after real-world deployment; and (3) explicitly training AI systems to be truthful via curated datasets and human interaction. A concerning possibility is that evaluation mechanisms for eventual truthfulness standards could be captured by political interests, leading to harmful censorship and propaganda. Avoiding this might take careful attention. And since the scale of AI speech acts might grow dramatically over the coming decades, early truthfulness standards might be particularly important because of the precedents they set.

Active reinforcement learning (ARL) is a variant on reinforcement learning where the agent does not observe the reward unless it chooses to pay a query cost c > 0. The central question of ARL is how to quantify the long-term value of reward information. Even in multi-armed bandits, computing the value of this information is intractable and we have to rely on heuristics. We propose and evaluate several heuristic approaches for ARL in multi-armed bandits and (tabular) Markov decision processes, and discuss and illustrate some challenging aspects of the ARL problem., Originally appeared at the NeurIPS 2016 "Future of Interactive Learning Machines (FILM)" workshop

AI systems are increasingly applied to complex tasks that involve interaction with humans. During training, such systems are potentially dangerous, as they haven't yet learned to avoid actions that could cause serious harm. How can an AI system explore and learn without making a single mistake that harms humans or otherwise causes serious damage? For model-free reinforcement learning, having a human "in the loop" and ready to intervene is currently the only way to prevent all catastrophes. We formalize human intervention for RL and show how to reduce the human labor required by training a supervised learner to imitate the human's intervention decisions. We evaluate this scheme on Atari games, with a Deep RL agent being overseen by a human for four hours. When the class of catastrophes is simple, we are able to prevent all catastrophes without affecting the agent's learning (whereas an RL baseline fails due to catastrophic forgetting). However, this scheme is less successful when catastrophes are more complex: it reduces but does not eliminate catastrophes and the supervised learner fails on adversarial examples found by the agent. Extrapolating to more challenging environments, we show that our implementation would not scale (due to the infeasible amount of human labor required). We outline extensions of the scheme that are necessary if we are to train model-free agents without a single catastrophe.

Providing Reinforcement Learning agents with expert advice can dramatically improve various aspects of learning. Prior work has developed teaching protocols that enable agents to learn efficiently in complex environments; many of these methods tailor the teacher's guidance to agents with a particular representation or underlying learning scheme, offering effective but specialized teaching procedures. In this work, we explore protocol programs, an agent-agnostic schema for Human-in-the-Loop Reinforcement Learning. Our goal is to incorporate the beneficial properties of a human teacher into Reinforcement Learning without making strong assumptions about the inner workings of the agent. We show how to represent existing approaches such as action pruning, reward shaping, and training in simulation as special cases of our schema and conduct preliminary experiments on simple domains., Presented at the NIPS Workshop on the Future of Interactive Learning Machines, 2016

Everyday social interactions are heavily influenced by our snap judgments about others’ goals. Even young infants can infer the goals of intentional agents from observing how they interact with objects and other agents in their environment: e.g., that one agent is ‘helping’ or ‘hindering’ another’s attempt to get up a hill or open a box. We propose a model for how people can infer these social goals from actions, based on inverse planning in multiagent Markov decision problems (MDPs). The model infers the goal most likely to be driving an agent’s behavior by assuming the agent acts approximately rationally given environmental constraints and its model of other agents present. We also present behavioral evidence in support of this model over a simpler, perceptual cue-based alternative., United States. Army Research Office (ARO MURI grant W911NF-08-1-0242), United States. Air Force Office of Scientific Research (MURI grant FA9550-07-1-0075), National Science Foundation (U.S.) (Graduate Research Fellowship), James S. McDonnell Foundation (Collaborative Interdisciplinary Grant on Causal Reasoning)